impact of the ocean’s overturning circulation on...

INTRODUCTION

The ocean contains 60 times as much carbon as theatmosphere and 17 times as much as the terrestrial biosphereincluding soils [Sarmiento and Gruber, 2006]. As such it isthought to exert a strong control on the concentration of CO2

in the atmosphere, and hence on climate, on time scales ofcenturies to thousands of years. The ocean affects atmos-pheric CO2 through physical and biological processes.Carbon-rich cold water (due to greater solubility at lower

temperatures) sinks at high latitudes and fills the deepocean, leading to larger carbon concentrations at depths.This “solubility pump” accounts for 30-40% of the surfaceto deep gradient of dissolved inorganic carbon (DIC)(Toggweiler et al., 2003; see also below e.g. lower left panelin Figure 2). Biological activity removes carbon from thesurface along with nutrients in the form of sinking particu-late organic matter (POM) and calcium carbonate (CaCO3)shells which is remineralized back to inorganic matter ordissolves in the abyss. This “biological pump” accounts for60-70% of the surface to deep DIC gradient. Sinking ofCaCO3 leads to a vertical gradient in alkalinity. Its produc-tion, though changes in the carbonate ion chemistry,increases the partial pressure PCO2 of CO2 in surface watersand hence atmospheric CO2. Both the solubility as well asthe biological pump are affected by the ocean’s circulation.

Impact of the Ocean’s Overturning Circulation on Atmospheric CO2

Andreas Schmittner

College of Oceanic and Atmospheric Sciences, Oregon State University, Corvallis, Oregon, USA.

Edward J. Brook and Jinho Ahn

Department of Geosciences, Oregon State University, Corvallis, Oregon, USA.

A coupled climate-carbon cycle model and ice core CO2 data from the last glacialperiod are used to explore the impact of changes in ocean circulation on atmos-pheric CO2 concentrations on millennial time scales. In the model, stronger winddriven circulation increases atmospheric CO2. Changes in the buoyancy drivendeep overturning in the Atlantic affect atmospheric CO2 only indirectly throughtheir effect on Southern Ocean stratification. In simulations with an abrupt andcomplete shutdown of the Atlantic overturning, stratification in the Southern Oceandecreases due to salinification of surface waters and freshening of the deep sea.Deeper mixed layers and steeper isopycnals lead to outgassing of CO2 in theSouthern Ocean and hence gradually increasing atmospheric CO2 concentrationson a multi-millennial time scale. The rise in CO2 terminates at the time of rapidresumption of deep water formation and warming in the North Atlantic, and CO2

levels subsequently gradually decrease. These model responses and a strong corre-lation between simulated atmospheric CO2 and Antarctic surface air temperatureswith little or no time lag are consistent with newly synchronized ice core data fromthe last ice age. Sensitivity experiments reveal that the amplitude of the response ofatmospheric CO2 is sensitive to the model background climatic state and decreasesin a colder climate owing to smaller changes in the overturning.

315

Ocean Circulation: Mechanisms and ImpactsGeophysical Monograph Series 173Copyright 2007 by the American Geophysical Union10.1029/173GM20

GM01073_CH20.qxd 9/8/07 5:26 PM Page 315

Models

How exactly do changes in ocean circulation affect atmos-pheric CO2 concentrations? A number of previous modelingstudies have addressed this question with different and some-times conflicting results. Siegenthaler and Wenk [1984] find adecrease of CO2 if the deep overturning circulation is increasedin their 4 box ocean-atmosphere model. In the similar model ofSarmiento and Toggweiler [1984] the response of atmosphericCO2 to changes in the deep overturning circulation dependsqualitatively on the exchange between the surface and deepocean at high latitudes. If this exchange is high or intermediate,increased circulation leads to lower CO2; if it is low, CO2

increases. Both models assume invariant and near zero surfacenutrient concentrations at low latitudes but variable concentra-tions at high latitudes. Both models are sensitive to high latitudeprocesses and predict an increase of CO2 if vertical mixing inthe Antarctic ocean is larger. In an idealized, albeit 3 dimen-sional, ocean model setup, Toggweiler et al. [2006] recentlydemonstrated that large self sustained atmospheric CO2 oscilla-tions of 30-45 ppmv can result from fluctuations in SouthernOcean stratification induced by changes in the wind stress.

Marchal et al. [1998, 1999], using a zonally averagedocean model, found that CO2 increased in transient simula-tions after North Atlantic Deep Water (NADW) formationwas shut down by freshwater input to the North Atlantic.They suggested decreased solubility of CO2 in warmer sur-face waters in the Southern Ocean as the main explanationfor higher atmospheric CO2. Contrary to those results othermodels predict an increase of CO2 with stronger NADW for-mation [Keir, 1988; Heinze and Hasselmann, 1993; Schulz et al., 2001; Köhler et al., 2005a, 2006].

Studies with stand alone terrestrial vegetation models[Scholze et al., 2003; Köhler et al., 2005b] find higher atmos-pheric CO2 due to reduced carbon storage on land broughtabout by climate changes associated with a reduction of thedeep Atlantic overturning circulation.

Observations

Observations of CO2 variations obtained by measuring airtrapped in bubbles of ancient ice from Antarctica [Indermühleet al., 2000], provide a unique opportunity to test carboncycle models on multi-millennial time scales. Figure 1 showsa new synchronization of these CO2 measurements based onCH4 correlation between Greenland and Taylor Dome,Antarctica [Ahn and Brook, 2007]. Due to the low resolutionof the Taylor Dome CH4 record the age model takes also intoaccount newly measured high-precision Byrd CO2 recordsbetween 47 and 65 ka [Ahn and Brook, 2007] and previouslypublished CO2 records from the Byrd ice core from 30 to 46ka [Neftel et al., 1988].

The following important features are apparent in theseobservations and robust with respect to dating uncertainties:(1) gradually increasing CO2 concentrations during some(but not all) cold times (stadials) in Greenland. Throughoutthis paper we also assume that stadials are associated withreduced deep overturning in the North Atlantic, adopting pre-vious hypotheses [Broecker et al., 1985] and numerous data(see e.g. the review paper by Broecker in this volume) andmodel studies [e.g. Schmittner et al., 2002, 2003] althoughthis view is not uncontested (see article by Wunsch in thisvolume). The Greenland stadials associated with CO2

increase also correspond to events of widespread ice raftingin the North Atlantic (Heinrich events 4, 5, 5a, and 6). (2)CO2 concentrations peak at the same time (within the datinguncertainty) of large rapid warming events in Greenland thatsignal an abrupt resumption of a vigorous overturning circu-lation. (3) After this transition CO2 gradually decreases dur-ing prolonged warm (interstadial) episodes in Greenland. (4)A strong correlation exists between surface air temperaturesover Antarctica and atmospheric CO2. (5) Antarctic tempera-tures and atmospheric CO2 vary on a multi-millennial timescale and do not show the abrupt transitions and higher fre-quency variations present in the Greenland temperature andmethane records. Longer stadials in Greenland are associatedwith larger amplitude warming events in Antarctica (EPICACommunity Members, 2006). Any successful theory or modelsimulation needs to reproduce the five features listed above.

In summary, ice core data suggest a correlation betweenocean circulation and atmospheric CO2 on millennial timescales whereas results from previous modeling appear conflict-ing and inconsistent and there is no generally accepted theoryas to whether atmospheric CO2 should increase or decrease dueto a stronger ocean circulation. In addition, all previous modelstudies were carried out with highly simplified models (eitherbox models, or zonally averaged ocean circulation models) thatdid not incorporate important physical (such as three dimen-sional deep and intermediate water circulation in the SouthernOcean or dynamical sea ice) or biological (such as the explicitrepresentation of phytoplankton) processes. Here we try to illu-minate the effect of both changes in the ocean’s wind and buoy-ancy driven overturning circulations on atmospheric CO2 usinga detailed 3-dimensional model of ocean circulation, ecosys-tem dynamics and carbon cycling embedded in a coupled cli-mate model of intermediate complexity including aninteractive dynamic terrestrial vegetation component.

A WORKING HYPOTHESIS

Let us, as a working hypothesis, assume that increasing theocean circulation leads to higher atmospheric CO2. Thishypothesis can be rationalized by the following thought exper-iment. Consider that biological activity always removes nutri-

316 OVERTURNING OCEAN CIRCULATION AND ATMOSPHERIC CO2

GM01073_CH20.qxd 9/8/07 5:26 PM Page 316

ents and carbon from the ocean surface due to the sinking ofparticulate organic matter. Plankton productivity is maintainedonly because the ocean circulation returns remineralized inor-ganic nutrients and carbon to the sunlit surface. Thus, in thelimit of a motionless ocean, surface waters would be stripped ofall nutrients, and DIC concentrations (and hence atmosphericCO2) would be low. In the limit of an infinitely strong circula-tion the deep and surface waters would be well mixed and sur-face nutrients, and DIC, and atmospheric CO2 would be high.In other words: in this case the biological and solubility pumpswould not be able to establish any vertical DIC gradient.

The atmospheric CO2 concentration in such a world of awell mixed ocean can easily be calculated as the partialpressure of CO2 (PCO2) in equilibrium with sea water with itspreindustrial global mean temperature (4°C), salinity (34.9),DIC (2.304 mmol/m3) and alkalinity (2.417 mmol/m3)concentrations as 518 ppmv (1 ppmv PCO2 = 2.13 gigatons ofcarbon [GtC]). This is almost twice the preindustrialatmospheric CO2 concentration of 280 ppmv and presents an

upper limit of the effect of ocean circulation on atmosphericCO2. This value is consistent with model experiments in whichthe solubility and biological pumps were switched off[Cameron et al., 2005]. Our hypothesis assumes finite plank-ton growth rates as well as a constant terrestrial carbon pool.The term “ocean circulation” refers to any vertical waterexchange, either by convection, eddy activity, or the large scaleoverturning circulation. Thus, our hypothesis suggests a simplerelationship between ocean circulation and atmospheric CO2,such that a stronger circulation leads to higher CO2.

Below we will present model experiments designed to testthis hypothesis and to quantify the effect of changes in oceancirculation on atmospheric CO2 concentrations. After adescription of the model in the following section, we presentexperiments in which the wind driven circulation and the buoy-ancy driven circulation (both through changes in surface fluxesand interior mixing) have been changed and the response of theglobal coupled carbon cycle has been analyzed in order tounderstand the simulated changes in atmospheric CO2.

SCHMITTNER ET AL. 317

Figure 1. Comparison of atmospheric CO2 and climate changes during the last ice age from Ahn and Brook (2007). Open square, CH4

from Byrd ice core, Antarctica [Blunier and Brook, 2001]; open circle, CO2 from Taylor Dome, Antarctica [Indermühle et al., 2000]; blackline, δ18Oice, as a proxy of surface temperature, from Byrd station, Antarctica [Johnsen et al., 1972] on the timescale of Blunier and Brook[2001]; gray line, δ18Oice from Greenland Ice Sheet Project 2 (GISP2) ice core (Summit, Greenland) [Grootes et al., 1993]. CO2 from TaylorDome is synchronized with GISP2 age [Ahn and Brook, 2007] using CH4 from Taylor Dome [Brook et al., 2000] and CO2 from Byrd station[Ahn and Brook, 2007; Neftel et al., 1988]. Vertical lines denote the peak of Antarctic warm events A1-A4 which are associated with rapidtransitions from a cold stadial to warm interstadial state in Greenland. Gray shaded bars represent Heinrich events [Sarnthein et al., 2001;Rashid et al., 2003]. Dansgaard/Oeschger warm events 8, 12, 14 and 17 and Heinrich events 4, 5, 5a and 6 are denoted by numbers. Notethat following these stadial to interstadial transitions atmospheric CO2 decreases gradually on a multi-millennial time scale.

GM01073_CH20.qxd 9/8/07 5:26 PM 17

MODEL DESCRIPTION AND ANALYSIS

The physical model is based on the University of VictoriaEarth System Climate Model [Weaver et al., 2001] version2.7. It includes a global, three dimensional, primitive equa-tions, ocean model with diffusion along and across isopyc-nal surfaces, a parameterisation of tracer advection due toeddies [Gent and McWilliams, 1990] and a tidal mixingscheme [Simmons et al., 2004]. This scheme leads toenhanced vertical mixing over rough topography and resultsin low diapycnal mixing in the pelagic pycnocline withdiapycnal diffusivities Kv there equal to the background dif-fusivity Kb = 2 × 10−5m2s−1. A simple, two dimensionalenergy balance model of the atmosphere is used with a pre-scribed seasonal cycle of winds as well as an interactivedynamical terrestrial vegetation/carbon cycle model[Meissner et al., 2003]. Of particular importance in thisstudy is a state of the art dynamic-thermodynamic sea icemodel which strongly improves stratification and deep andintermediate water formation in the Southern Ocean[Saenko et al., 2002].

The marine ecosystem model is an improved version ofSchmittner et al. [2005] with a parameterization of fast nutri-ent recycling due to microbial activity after Schartau andOschlies [2003]. It considers two phytoplankton classes(nitrogen fixers and other phytoplankton), one zooplanktonclass, sinking particulate organic matter (detritus) as wellas interactive cycling of nitrogen, phosphorus and oxygenincluding denitrification in low oxygen waters [Schmittner et al., in press]. The inorganic variables include oxygen (O2),two nutrients, nitrate (NO3) and phosphate (PO4) as well asDIC and total alkalinity (ALK). They are linked throughexchanges with the biological compartments via Redfieldstoichiometry (RP:N = 1/16, RO:N = 13, RC:N = 7).

Calcium carbonate (CaCO3) production is parameterizedas a fixed ratio (rCaCO3:C = 0.028) of the net production ofPOM in the water column. Because of nutrient and carbonrecycling within the euphotic zone (regeneration), net pro-duction of POM is always larger than export production ofPOM out of the euphotic zone. We assume that CaCO3 isnot recycled within the euphotic zone but immediatelyexported because CaCO3 sinks much faster than POM. Theparameter rCaCO3:C has been tuned to reproduce the globalvertical ALK gradient. Thus the rain ratio of calcite toorganic matter flux at the base of the euphotic zone (~100m)is variable in our model and depends e.g. on nutrientcycling within the euphotic zone. As shown in Schmittner etal. (submitted manuscript) its values and latitudinal distrib-ution (are consistent with recent observational estimates[Sarmiento et al., 2002]. Dissolution of CaCO3 is deter-mined by instantaneous sinking with an e-folding depth of3500m.

Formulations of air-sea gas exchange and carbon chem-istry follow protocols from the Ocean Carbon-Cycle ModelIntercomparison Project [OCMIP, Orr et al., 1999] asdescribed in Ewen et al. [2004]. One deviation from theOCMIP protocol is the treatment of the effect of freshwaterfluxes on surface DIC and ALK. Here we follow theapproach of Marchal et al. [1998] and use a salinity normal-ized tracer, e.g.: , where DICm is themodel value without dilution/concentration by surface fresh-water fluxes and DIC the actual in situ value, S is the localsalinity and is a reference salinity. In the calculationof air-sea gas exchange and surface chemistry as well as forcomparison with observations DIC is used. This methodavoids the use of virtual freshwater fluxes at the surface butsensitivity tests showed that resulting surface PCO2 and air-sea fluxes are very similar for both methods.

A more complete description of the model as well as adetailed comparison with present day observations will begiven elsewhere (Schmittner et al., submitted manuscript.).There it is also shown that the model is consistent with alarge array of observations including radiocarbon (bomb andnatural), anthropogenic carbon and chlorofluorocarbons, aswell as air-sea fluxes of CO2. For example, the model ver-sions used here are consistent with all metrics proposed byMatsumoto et al. [2004].

Here, we use two model versions termed wNPs and sNPsin Tables 1 and 2. Model version wNPs exhibits a weak NorthPacific stratification caused by an underestimated halocline.In order to correct this systematic bias model version sNPs(strong North Pacific stratification) was constructed byadding 0.1 Sv of freshwater permanently to the North Pacificnorth of 40°N (a compensating flux in the rest of the worldocean was used in order to conserve global salinity). Thisflux correction has the desired effect of increasing the strati-fication and improves tracer distributions such as oxygen andnutrients in the North Pacific (not shown); however, it is notclear whether or not such a permanent flux correction biasesthe transient model response. The use of both model versionscircumvents this problem. It is also motivated by the fact thatthe response of the North Pacific circulation to a perturbationof the Atlantic overturning depends on the stratification there[Schmittner and Clement, 2002]. During the glacial, stratifi-cation in the North Pacific was most likely different fromtoday. Stable isotope records from below 1 km depth suggestbetter ventilation of the upper ocean [Keigwin, 1998],whereas it was inferred from surface productivity recordsthat stratification in the glacial North Pacific was larger thantoday [Jaccard et al., 2005]. Here we use both model ver-sions in order to test the robustness and sensitivity of ourresults to different stratification in the North Pacific.

Atmospheric CO2 is calculated fully prognostically in onewell-mixed atmospheric box and its changes interactively

S =34 9.

DIC DIC S Sm = ⋅ /


GM01073_CH20.qxd 9/8/07 5:26 PM Page 318

affect the radiative balance and hence climate. Due toremaining model imperfections the simulated preindustrialatmospheric CO2 concentration is 308 ppmv, slightly higherin both model versions than the observed value of ~280 ppmv.The simulated oceanic carbon inventories, however, as well asthe 3-dimensional patterns of dissolved inorganic carbon andalkalinity distributions are consistent with observational esti-mates as will be shown below.

In order to separate the contributions of the biological andsolubility pump we use a model version with an abiotic oceancarbon cycle only (termed NoBio in the following). In thismodel version surface alkalinity is fixed at its observedglobal mean value of mol/m3. The lower leftpanel in Figure 2 shows that in our model the solubility pumpaccounts for only 1/3 of the surface to deep DIC gradient,whereas the biological pump accounts for 2/3.

To our knowledge this is the most complete and detailedmodel yet applied to study the impact of ocean circulationchanges on atmospheric CO2 on millennial time scales.However, it needs to be clearly stated that certain processesare still highly simplified. Atmospheric heat and moisturetransports, for instance, are not influenced by changes inwind velocities. Wind stress to the ocean and sea ice is pre-scribed as well and does not respond to climatic changes. Wewill address some of these model simplifications and theirpossible influences on reported results and conclusionsbelow and in the discussion section. Most of our simulationshave been performed for a preindustrial background climate.However, we do include one idealized experiment using acolder, more glacial background climate in a sensitivity test.

Analysis of Ocean Circulation and PCO2 Changes

Table 1 presents indices of the large scale circulation in dif-ferent model simulations close to equilibrium, whereas Table 2lists important variables of the carbon cycle in those simula-tions. The atmospheric concentration of CO2 is strongly con-trolled by the sea surface PCO2. This can be seen in similar(albeit not exactly the same) changes of atmospheric CO2 andsurface ocean PCO2 in the different experiments listed inTable 2. Differences between atmospheric CO2 and sea sur-face PCO2 are caused by changes in air-sea equilibration, e.g.due to changes in sea ice cover or changes in the residencetime of surface waters. In the model, sea surface PCO2 dependsonly on temperature T, salinity S, DIC and ALK: PCO2 = PCO2

(T,S,DIC,ALK), however non-linearly. This allows decompo-sition of the simulated total changes of PCO2 into individualcontributions [Marchal et al., 1998]. In Table 2 we show PCO2

changes brought about by variations of T and S that deter-mine primarily physical solubility of surface waters ∆PCO2

(T,S) separately from changes brought about by redistribu-tions of DIC and ALK which involve changes in biological

2 36775. /⋅ S S

and physical cycling ∆PCO2(D,A). For example, the solubilityeffect is calculated as ∆PCO2(T,S) = PCO2(T(EXP),S(EXP),DIC(CTR),ALK(CTR)) – PCO2(T(CTR),S(CTR),DIC(CTR),ALK(CTR)), where EXP denotes the equilibrium at the end ofthe experiment and CTR denotes the end of the correspondingcontrol run. Generally, PCO2 changes caused by changes in Sare small, thus the variations in PCO2(T,S) reported in Table 2are mainly caused by changes in T. Changes due to DIC aregenerally of opposite sign as those due to ALK [Marchal et al., 1998]. Note that due to non-linearities in the PCO2 equa-tion, the sum of the changes due to T and S, and the changesdue to DIC and ALK does not need to add up exactly to thetotal PCO2 change.

WIND DRIVEN CIRCULATION

In this section two idealized experiments are discussed, inwhich the wind stress everywhere at the surface of the oceanand sea ice was halved (0.5 × WS) or doubled (2 × WS). Theshallow wind driven overturning circulation respondsapproximately linearly to these changes in the forcing. In thecontrol run tropical Ekman cells in the northern (40 Sv) andsouthern (54 Sv) hemisphere amount to equatorial upwellingof 94 Sv. In the Pacific between 9°N and 9°S 42 Sv upwellacross 50m into the model surface layer consistent withobservational estimates of 41 ± 6 Sv [McPhaden and Zhang,2002]. In experiment 0.5 × WS the tropical overturning cellsare reduc ed to 22 Sv and 28 Sv, respectively, and in experi-ment 2 × WS equatorial upwelling is increased to 193(= 90 + 103) Sv. The Antarctic Circumpolar Current (ACC)accelerates from 80 Sv in the control run “wNPs ctr” to 147Sv in 2 × WS and decelerates to 54 Sv in 0.5 × WS. TheIndonesian Throughflow, with 19 Sv in the control run con-sistent with observational estimates of 16 ± 5 Sv [Ganachaudand Wunsch, 2000], slows to 13 Sv in 0.5 × WS and speedsup to 27 Sv in 2 × WS.

Increasing the wind stress leads not only to a more vigor-ous upper ocean circulation but it increases deep overturningas well (Table 1). Upwelling in the Southern Ocean increasesas does the rate of formation and export of NADW. Theseresults are consistent with earlier findings [Toggweiler andSamuels, 1995] suggesting increased wind stress over theSouthern Ocean leads to more Ekman suction of water fromthe deep ocean and thus to intensified NADW formation. Inour case, additionally to this effect a more vigorous low lati-tude Ekman circulation increases subduction of warm sur-face water into the thermocline and leads to a deepening ofthe pycnocline at low latitudes which intensifies the merid-ional pressure gradients and thus accelerates the buoyancydriven deep overturning as well.

This acceleration of the surface and deep flows leads tofaster nutrient delivery to the surface and thus has a strong


GM01073_CH20.qxd 9/8/07 5:26 PM Page 319


Figure 2. Globally horizontally averaged profiles of phosphate, natural radiocarbon, temperature, salinity, DIC and ALK. Symbolsshow present day observations from the World Ocean Atlas 2001 [Conkright et al., 2002] and preindustrial estimates from GLODAP[Key et al., 2004]. Solid line: present day control run (wNPs ctr), dashed line: 0.5 × WS, dotted line: 2 × WS. The dash-dotted line inthe lower left panel shows the preindustrial DIC profile of the inorganic model (NoBio).

GM01073_CH20.qxd 9/8/07 5:26 PM Page 320


Table 1. Circulation indices from equilibrium experiments. Columns show the Antarctic Circumpolar Current (ACC) flow through DrakePassage (68°W), Indonesian Throughflow (ITF), Northern and Southern Tropical Ekman Cells (NTEC, STEC) computed as themaximum/minimum global streamfunction (vertically and zonally integrated meridional velocity) between 20°S-20°N, maximum ofOverturning in North Atlantic (ONA) computed as the maximum streamfunction in the North Atlantic below 300m, export of North AtlanticDeep Water into the Southern Ocean (NADW) computed as maximum Atlantic streamfunction at 35°S below 300m, Overturning in NorthPacific (ONP) computed as the maximum North Pacific streamfunction below 300m, CircumPolar Deep Water (CPDW) flux into Indo-Pacific computed as minimum Indo-Pacific streamfunction at 35°S below 300m, UpWelling in the Southern Ocean (UWSO) computed asthe maximum streamfuction south of 35°S below 300m, Antarctic DownWelling (AADW) computed as the minimum streamfunction southof 60°S below 300m, globally integrated Antarctic Bottom Water (AABW) cell computed as the minimum streamfunction below 2000m.

Model ACC ITF NTEC STEC ONA NADW ONP CPDW UWSO AADW AABW Version Experiment (Sv) (Sv) (Sv) (Sv) (Sv) (Sv) (Sv) (Sv) (Sv) (Sv) (Sv)

wNPs ctr 80 19 40 54 15 14 1 11 22 4 140.5 × WS 54 13 22 28 12 10 2 12 7 3 142 × WS 147 29 90 103 21 19 2 12 49 2 15NADW off 79 12 39 55 0 0 9 8 19 4 13NADW off + 79 11 38 50 0 0 11 8 18 4 13∆τGENESIS

KvSO = 1 71 19 39 49 18 17 0 12 21 3 9LGM ctr 88 17 41 54 13 11 2 9 21 7 14LGM NADW 86 13 41 53 2 2 7 8 19 8 14off

sNPs ctr 80 20 40 54 17 16 0 12 22 3 12NADW off 84 14 38 56 0 0 2 8 17 3 13tidal off 83 17 40 52 15 14 0 8 22 3 9

Table 2. Global carbon and productivity results from equilibrium experiments. Columns show atmospheric CO2 concentration; difference inatmospheric CO2 from control run (∆CO2); changes in ocean (∆CO) and land (∆CL) carbon inventories; surface ocean partial pressure of CO2

(PCO2) over sea ice free area; changes in PCO2 (∆PCO2) can be separated into changes in solubility due redistributions of temperature andsalinity ∆PCO2(T,S) and changes involving biology due to redistributions of DIC and ALK ∆PCO2(D,A). Export Production (EP) throughsinking of particulate organic matter, New Production (NP), Net Primary Production (NPP)

Model

∆PCO2 ∆PCO2

Version Experiment

CO2 ∆CO2 ∆CO ∆CL PCO2 ∆PCO2 (T,S) (D,A) EP NP NPP

ppmv ppmv GtC GtC ppmv ppmv ppmv ppmv GtC/yr GtC/yr GtC/yr

wNPs ctr 308 0 0 0 295 0 0 0 6.2 7.8 530.5 × WS 284 –24 +109 –59 270 –25 –8 –20 5.0 6.0 412 × WS 389 +81 –364 +194 380 +85 +22 +67 9.2 12 90NADW off 336 +27 –104 +46 322 +27 +8 +20 5.8 7.3 51NADW off + 336 +27 –104 +45 322 +30 +9 +22 5.8 7.4 51∆τGENESIS

KvSO = 1 326 +21 –89 +45 313 +21 +5 +17 7.1 8.8 61NoBio ctr 295 0 0 0 285 0 0 0 NA NA NANoBio NADW 290 –5 28 –16 278 –7 –1 –6 NA NA NAoffLGM ctr 281 0 0 0 261 0 0 0 5.5 6.9 40LGM NADW 286 +5 –18 +8 266 +5 +1 +4 4.9 6.2 37off

sNPs ctr 308 0 0 0 298 0 0 0 6.2 7.8 50NADW off 332 +26 –71 +26 318 +20 +6 +15 5.6 7.2 50tidal off 286 –22 +103 –49 272 -26 -11 –18 5.4 6.9 46

GM01073_CH20.qxd 9/8/07 5:26 PM Page 321

impact on the ecosystem and its productivity (Table 2). Newproduction increases by more than 50% and net primary pro-duction by about 70% in experiment 2 × WS compared withthe control run. Surface nutrient concentrations increase andthe nutricline weakens (Figure 2). Along with nutrients, car-bon is shifted from the deep and intermediate ocean to thesurface, where part of it escapes to the atmosphere. Surfaceocean PCO2 changes are dominated by redistributions of DICand ALK whereas changes due to T and S are smaller but notnegligible (Table 2).

Atmospheric CO2 levels are consequently higher thestronger the wind stress (Table 2), consistent with our work-ing hypothesis. However, the response of CO2 is nonlinearsuch that increasing the wind stress has a stronger effect thanits reduction. The reason for this asymmetry is presumablyrelated to the fact that surface nutrient concentrations arealready very low in most parts of the tropical and subtropicaloceans (globally averaged surface PO4 is 0.55 mmol/m3 inthe observations and 0.58 in the wNPs control run). Thusreduced upwelling in experiment 0.5 × WS does not affectmuch nutrient concentrations in these waters and only leadsto a small reduction in globally averaged surface phosphateto 0.52 mmol/m3, whereas increased upwelling in run2 × WS strongly increases surface nutrient concentrationseverywhere to 1.0 mmol/m3. This also explains the asymmet-ric response of productivity (Table 2). The strong increase inproductivity and nutrient cycling in the surface layers inexperiment 2 × WS also causes an acceleration of the car-bonate pump (visible as the larger vertical gradient of alka-linity in Figure 2 since production of CaCO3 is parameterizedas a constant fraction of POM production. This mechanismcontributes to the rise in PCO2 of surface waters and henceatmospheric CO2.

BUOYANCY DRIVEN CIRCULATION

Simulating a Shutdown of the Atlantic Overturning

In order to investigate the role of the buoyancy drivenocean circulation on atmospheric CO2 an experiment wasperformed, in which deep water formation in the NorthAtlantic was stopped abruptly through the application of afreshwater pulse (Figure 3). This experiment is motivated bythe paleo record from the last glacial period, which, as dis-cussed in the introduction, indicates that such large andabrupt changes in ocean circulation have indeed occurredand that they were related to variations in atmospheric CO2

[Indermühle et al., 2000].In response to the rapid reduction in ocean circulation

atmospheric CO2 increases gradually in our model on a mul-timillennial time scale. This is, at least at first glance, con-trary to our working hypothesis. In the following we present

a detailed analysis in order to better understand this unex-pected model behavior. First we notice that the atmosphericCO2 increase is caused by a reduction in the oceanic carboninventory by more than 100 GtC (Figure 3, Table 2), whereasthe terrestrial carbon pool acts as a buffer increasing by 46GtC. Less than one third (14 GtC) of this increase is due tochanges in the soil carbon pool whereas two thirds (32 GtC)are caused by vegetation carbon changes. Vegetation carbondeclines in northern and central Europe and increases at lowlatitudes and around the North Pacific (Plate 1).

Decreased solubility due to slightly warmer sea surfacetemperatures contributes 8 ppmv (Table 2) to the totalincrease of surface water PCO2 of 27 ppmv, but the dominantfactor in explaining the increase in atmospheric CO2 is ratherredistribution of DIC and ALK. The net changes in theoceanic carbon inventory are the result of much larger redis-tributions between the different ocean basins (Table 2,Figures 3 and 4). The large interbasin gradients of DIC andALK in the control simulation are erased if NADW isstopped (Fig 4). Whereas the Atlantic inventory increases byalmost 300 GtC, carbon storage in the other ocean basinsdeclines. The largest reduction occurs in the Pacific with 340GtC in model version wNPs and 289 GtC in model sNPs. Thelarger decrease in the Pacific carbon inventory in modelwNPs is due to a shallow overturning circulation that devel-ops in the North Pacific (Table 1) in response to the decreaseof the Atlantic overturning consistent with the Atlantic-Pacific seesaw mechanism [Saenko et al., 2004]. This over-turning, which is restricted to the northern hemisphere,removes carbon from the upper ocean in the North Pacific bydownwelling of low nutrient, low carbon surface waters tointermediate depths of about 1500m. In model version sNPssubduction in the North Pacific is suppressed due to strongerstratification (Table 1) and thus the carbon loss from thePacific is smaller.

In the abiotic run atmospheric CO2 decreases slightly (by~5 ppmv, Figure 3B, Table 2) and changes in the individualocean basins are much smaller than in the full model (Table 3,thin lines in Figure 4). These results support the conclusionthat changes in biologically mediated DIC and ALK arerequired to explain the full model response and that solubil-ity changes play only a secondary role, contrary to thehypothesis of Martin et al. [2005].

Globally averaged DIC concentrations decrease below~600m depth, with a maximum around 1300m (solid lines inFigure 5). This mid-depth maximum of depleted DIC con-centrations is a robust feature in both model versions (wNPsand sNPs, not shown). Alkalinity shows a similar depletionaround 1300m. Globally deep water freshens by 0.3 salinityunits due to the missing injection of salty NADW, which alsoleads to a salinification of surface waters by ~0.2 units.Reduced upwelling of cold water into the thermocline allows


GM01073_CH20.qxd 9/8/07 5:26 PM Page 322

Figure 3. Transient model response to a shutdown of the Atlantic Overturning (dashed line in A) forced by a pulse of freshwater inputto the North Atlantic (thin solid line in A). (B) Atmospheric CO2 concentration. (C) Change in land and ocean carbon inventories. (D)Surface air temperature (SAT) in Greenland. (E) Global SAT (dashed, left scale) and SAT in Antarctica (solid, right scale). (F) Changein carbon inventories in the different ocean basins (Indian: dotted; Atlantic: short dashed; Pacific: short-long dashed; Southern Ocean:long dashed; Global: solid). Lines with square symbols in (B) and (C) represent the model version without ocean biology (NoBio)

GM01073_CH20.qxd 9/8/07 5:26 PM Page 323


Plate 1. Vegetation (top) and soil (bottom) carbon pools in the preindustrial control simulation “wNPs ctr” (left panels) and their changesas a response to a permanent shutdown of the Atlantic overturning circulation (“wNPs NADW off ” minus “wNPs ctr”; right panels).

GM01073_CH20.qxd 9/8/07 5:26 PM Page 324


Figure 4. Basin wide averaged profiles of DIC (left) and ALK (right) versus depth. The thick solid line represents the controlsimulation “wNPs ctr” (year 0 in Figure 3), the thick dashed line the equilibrium without NADW formation “wNPs NADW off ” (year3700 in Figure 3). The Southern Ocean is defined south of 40°S. Thin lines show the corresponding profiles from the abiotic modelversion (NoBio). The gray shaded area shows observations with published error estimates (Key et al., 2004). In the case of DIC,anthropogenic CO2 was subtracted to yield preindustrial values.

GM01073_CH20.qxd 9/8/07 5:26 PM Page 325

increased downward diffusion of heat at low latitudes andhence leads to warming of the upper ocean. Nutrient concen-trations decrease in the upper 2 km and increase below. Thisis consistent with a vertical nutrient shift and reduced pro-ductivity associated with weaker deep water formation in theNorth Atlantic reported earlier [Schmittner, 2005].

Apparent oxygen utilization (AOU = sat(O2) - O2, the dif-ference between the temperature dependent oxygen satura-tion and the in situ O2 concentration), decreases in the deepocean by up to 90 mmol/m3. This corresponds to a ~30%reduction and is consistent with a decrease of the sinking andremineralization of POM into the deep ocean by a similaramplitude (not shown). This reduction of AOU implies adecrease of deep ocean remineralized phosphate (Pr = AOU/RO:P, with RO:P = 208 being the Redfield ratio of oxygen tophosphorus used in the model) by about 0.4 mmol/m3. Usinga carbon to phosphorus ratio of RC:P = 112 this correspondsto a decrease of DIC by about 40 mmol/kg. Thus, thedecrease of mid-depth carbon and phosphate concentrationscan be attributed to a reduction of remineralized matter,whereas below 2 km depth larger increases in preformednutrients Pp = PO4 - Pr (and carbon) dominate the inorganictracer changes.

The surface to deep density difference decreases globallyby about 0.2 kg/m3, mainly due to the throttled subduction ofrelatively salty NADW to the deep ocean. In the SouthernOcean this is apparent by the absence of the tongue of highsalinity NADW at mid depths (Plate 2) which leads to a largedecrease in stratification (the average difference between sur-face and deep potential density south of 40°S decreases by ~25%, Figure 6) along with deepening of the surfacemixed layer. Particularly around 58°S in the Pacific sector(not shown) maximum mixed layer depths increase dramati-cally tapping into the high DIC waters of the abyssal ocean.At this latitude the 27.6 σΘ isoline comes close to the surface

in experiment “wNPs NADW off ” (Plate 2). This isopycnalsurface deepens at low latitudes to about 1500m. Thus, betterventilation of deep water around this potential densitythrough steeper isopycnals and deeper mixed layers in theSouthern Ocean around 58˚S provides a consistent explana-tion for the depletion of DIC concentrations around 1500mdepth (Figure 5). The slow process of along isopycnal diffu-sion is also consistent with the long multi-millennial timescale of the simulated atmospheric CO2 response.

Our analysis indicates that weakened stratification in theSouthern Ocean is ultimately responsible for lower deepwater DIC concentrations and higher atmospheric CO2 in thesimulations without NADW formation. This result is consis-tent with earlier studies emphasizing the importance ofSouthern Ocean ventilation on atmospheric CO2

[Siegenthaler and Wenk, 1984; Sarmiento and Toggweiler,1984; Sarmiento and Orr, 1991; Toggweiler, 1999; Sigmanand Boyle, 2000; Toggweiler et al., 2006].

The simulated anti-phase behavior of stratification in theNorth Atlantic and Southern Ocean also supports the hypoth-esis of a bipolar seesaw of deep water formation suggested byBroecker [1998]. Deeper winter mixed layers result inincreased heat loss to the atmosphere (not shown) and con-tribute to warming of the air at high southern latitudesthereby explaining the correlation between atmospheric CO2

and air temperatures around Antarctica.

Comparison With the Paleo Record

The paleo record from Marine Isotope Stage 3 (MIS3, 30-60 ka BP) during the last ice age displays most clearly sta-dial to interstadial transitions (Figure 1), whereas rapid cool-ing (interstadial-stadial) events are not so unambiguouslyidentifiable. Stadial-interstadial transitions are associatedwith rapid warmings of 8-15°C in Greenland [Huber et al.,2006], rapid increases in methane, and a resumption of theAtlantic overturning circulation. In order to improve compar-ison with the observations two additional model experimentswere conducted that include stadial-interstadial transitionsand differ in the duration of the stadial phase (Figure 7). Thesimulated temperature changes in Greenland are less than 5°Cand clearly underestimated. Missing atmospheric dynamicsand the coarse resolution might be responsible for this bias.Simulated atmospheric CO2 raises gradually after thetransition to the stadial phase. It peaks right at the stadial-interstadial transition and gradually decreases afterwards. Thesimulated atmospheric CO2 changes are highly correlatedwith temperature changes over Antarctica with little or notime lag. All of these features are consistent with the observa-tions (Figure 1).

The simulated amplitude of the CO2 variations depends onthe duration of the stadial phase. It is about 15 ppmv for a short


Table 3. Changes in ocean carbon inventory in the different oceanbasins.

Model Experiment

∆CO ∆CO ∆CO ∆CO ∆CO

Version

. total Atl. Pac. Ind. SO

GtC GtC GtC GtC GtC

wNPs NADW off –104 +288 –340 –26 –28NADW off + ∆τGENESIS –104 +330 –379 –25 –31KvSO = 1 –89 +4 –22 –13 –58NoBio NADW off +28 +46 –31 –2 +16LGM NADW off –18 +170 –187 –6 +6

sNPs NADW off –79 +310 –289 –49 –52tidal off +113 –23 +76 –4 +64

GM01073_CH20.qxd 9/8/07 5:26 PM Page 326


Figure 5. Globally averaged profiles of tracer changes caused by a cessation of NADW formation (“wNPs NADW off ” minus “wNPsctr”, solid lines) or by removing tidal mixing (“sNPs tidal off ” minus “sNPs ctr”, dashed lines). Bold lines represent the global ocean,thin lines the Southern Ocean south of 40°S. DICm and ALKm (mmol/m3) anomalies do not include the effects of dilution (i.e. are notnormalized with salinity in contrast to those shown in Plate 1) because dilution affects neither the global carbon inventory nor surfacePCO2 and thus does not impact atmospheric CO2. Dash-dotted lines in the top left panel show DIC anomalies due to a cessation ofNADW formation in the inorganic model (NoBio).

GM01073_CH20.qxd 9/8/07 5:26 PM Page 327


Plate 2. Zonally averaged DICm concentrations (mmol/m3) in the Southern Ocean (gray shading), annual mean isopycnal surfaces (σΘ, red lines), salinity (green lines, the 34.65 isohaline is thicker), and zonal and monthly maximum of mixed layer depth (calculatedas depth of 0.1 σΘ difference from surface layer, blue line).

GM01073_CH20.qxd 9/8/07 5:26 PM Page 328

(1000 year long) stadial and almost double (30 ppmv) for a2000 year long stadial. We hypothesize that this is due to thegradual nature of the CO2 change, which does not reach equi-librium for short stadial durations. We propose this as anexplanation of the lack of large high frequency variability inthe observed CO2 record. A stadial of less than 1000 yearduration would have almost no appreciable atmospheric CO2

signal owing to the sluggish response of the oceanic carboncycle dominated by slow isopycnal diffusion.

A “Glacial” Simulation

All experiments described above were performed with apre-industrial (interglacial) background climate. However,the climate during MIS3 was much different, with large icesheets covering North America and Europe and a cooler anddryer atmosphere in general. As a first idealized test of therobustness of our model results with respect to the back-ground climate, additional experiments have been conductedwithin a colder climatic state. A perturbation of 2.5 W/m2 hasbeen added permanently to the outgoing longwave radiationat the top-of-the-atmosphere which leads to a global coolingof surface air temperatures by 3°C. This corresponds approx-imately to conditions during the Last Glacial Maximum(LGM). This simulation is termed LGM ctr in Tables 1 and 2.However, no other changes (such as imposing ice sheets orlowering sea level) have been made, and as such this experi-ment does not represent a realistic glacial climate but ratheran idealized sensitivity experiment. Note also that atmos-pheric CO2 is still at an interglacial level of 280 ppmv in this experiment. Obviously the model is unable to simulate

the observed glacial-interglacial CO2 change of ~80-100ppmv. However, in this paper we are not concerned with theglacial-interglacial CO2 problem, which remains unsolved.

The export of North Atlantic Deep Water to the SouthernOcean is reduced by more than 20% in the LGM ctr run(Table 1) and downwelling along Antarctica is increased dueto enhanced sea ice formation. Both features have alreadybeen reported in earlier glacial studies with the same model[Weaver et al., 1998; Schmittner, 2003]. The response ofatmospheric CO2 to a shutdown of NADW formation (exper-iment “LGM NADW off ”) is strongly reduced from 27 ppmvin the preindustrial model to only 5 ppmv (Table 2).Redistribution of carbon between the different ocean basinsis qualitatively similar to the interglacial model version(Table 2) in that it shows a large increase in the Atlantic anda decrease in the Pacific; however, the magnitude is muchsmaller. Inspection of the vertical changes (not shown) alsoreveals qualitatively similar results to the interglacial model(e.g. the maximum decrease of DIC around 1300m depth inFigure 6) but with reduced amplitudes.

These results suggest that the responses of the oceaniccarbon cycle and atmospheric CO2 are sensitive to the back-ground climate, presumably owing to the weaker mean stateof the Atlantic overturning in the LGM ctr run. This reducesthe effect of NADW on the stratification in the SouthernOcean and leads to a muted response if NADW is shut off. The observed amplitude of atmospheric CO2 changes of~15 ppmv (Figure 1) during MIS3 is between the amplitudein the interglacial (27 ppmv) and the full glacial (5 ppmv)model simulations and as such it is not inconsistent with ourmodel simulations.


Figure 6. Depth profiles of potential density (left), salinity (center) and potential temperature (right) in the Southern Ocean for thecontrol run (“wNPs ctr”, solid lines), the equilibrium state without NADW formation (“wNPs NADW off ”, dashed lines) and presentday observations [Conkright et al., 2002, symbols].

GM01073_CH20.qxd 9/8/07 5:26 PM Page 329


Figure 7. Response of atmospheric CO2 (second panel from below) and SATs in Greenland (second panel from above) and Antarctica(bottom panel) to a collapse of the Atlantic overturning (thick lines in top panel) at year 0 and its resumption after 1000 years (solidlines) and 2000 years (dashed lines). Thin lines in top panel denote the freshwater forcing.

GM01073_CH20.qxd 9/8/07 5:26 PM Page 330

Additional Sensitivity Experiments

Influence of wind changes. In order to quantify the effectof wind driven ocean circulation changes brought about bythe response of atmospheric dynamics to the shutdown of theAtlantic overturning circulation offline simulations with theGENESIS atmospheric General Circulation Model (GCM)have been conducted using SST and sea ice boundary condi-tions from the UVic model runs “wNPs ctr” and “wNPsNADW off ”. Resulting wind stress anomalies have beenused to force an experiment otherwise equivalent to ‘wNPsNADW off ”. This experiment is labeled “wNPs NADWoff + ∆τ GENESIS” in Tables 1 and 2. The resulting wind stressanomalies and a more detailed description of the method aswell as its effect on subsurface oxygen concentrations isdescribed elsewhere (Schmittner et al., submitted manu-script). The resulting changes in the wind driven circulationhave no effect on atmospheric CO2 as evident by anunchanged equilibrium concentration (Table 2) and almostidentical transient CO2 changes (not shown).

Effect of Southern Ocean stratification. In order to furtherexplore the effect of Southern Ocean stratification on atmos-pheric CO2 concentrations we performed two additional sim-ulations. In the first experiment the enhancement ofdiapycnal mixing due to tidally induced energy dissipationover rough topography was set to zero everywhere in theocean. In this experiment (called “tidal off ” in Tables 1 and 2)the diapycnal diffusivity is constant everywhere and equal toits background value (Kv = Kb = 2 × 10−5m2/s). The simu-lated circulation changes are consistent with earlier results[Saenko and Merryfield, 2005] in that the global AntarcticBottom Water cell is strongly reduced in strength, lessCircumpolar Deep Water enters the Pacific, but the overturn-ing in the Atlantic is only marginally weaker (Table 1).Reduced topographically enhanced mixing in the SouthernOcean leads to increased stratification there (see dashed linesin Figure 5) and lower atmospheric CO2 by 22 ppmv.

Decreased ventilation of the deep ocean leads to older sur-face to deep radiocarbon age differences. Globally the sur-face to deep ∆14C gradient increases by 100 permil. Reducednutrient delivery to Southern Ocean surface waters leads to adecrease of global productivity and export of POM. However,despite slightly reduced remineralization of POM in the deepocean (by ~10%, not shown) AOU increases, indicatinghigher concentrations of remineralized nutrients Pr in thedeep sea. Thus, we conclude that these higher remineralizednutrient concentrations result from reduced removal of thesenutrients from the deep sea by weaker ventilation.

In a second experiment the vertical diffusivity in theSouthern Ocean south of 40°S and below 500m depth has beenset to a minimum of 1 cm2/s (“Kv

SO = 1” in Tables 1 and 2) in

accordance with present day observations [Naveira Garabatoet al., 2004; Sloyan, 2005] which show larger background val-ues than our default model. The reason for the higher back-ground diffusivities in the Southern Ocean is not clear butmight be related to interaction of the ACC with the roughtopography [Sloyan, 2005], an effect not considered in the tidalmixing scheme. Anyway, increased mixing in the SouthernOcean only weakens stratification there and leads to anincreased overturning in the North Atlantic (Table 1).Atmospheric CO2 increases by 21 ppmv (Table 2). Both exper-iments demonstrate that stratification in the Southern Ocean isimportant in controlling atmospheric CO2 and not the Atlanticoverturning circulation itself. Adding the effect of enhancedmixing due to tides and that of increased background mixing,the total effect of enhanced mixing in the present day SouthernOcean leads to more than 40 ppmv higher atmospheric CO2

concentrations. A more detailed description of the tracer dis-tributions in these experiments is planned elsewhere.

DISCUSSION AND CONCLUSIONS

Our model simulations suggest that changes in both windand buoyancy driven ocean circulation can have a largeimpact on atmospheric CO2 concentrations. Results fromtransient simulations with large and rapid changes of theAtlantic overturning circulation are consistent with a numberof characteristics in the glacial ice core record. After anabrupt collapse of the overturning, atmospheric CO2 slowlyincreases by more than 20 ppmv on a multi-millennial timescale. This time scale is set by the slow equilibration of thedeep ocean through along isopycnal mixing processes.Therefore, higher frequency (centennial) oscillations of theoverturning excite no measurable response in atmosphericCO2 and the amplitude of the CO2 variation is larger forlonger period oscillations.

Simulated Antarctic temperatures and atmospheric CO2

peak at the time of rapid resumption of the circulation andgradually decline afterwards. Changes in stratification in theSouthern Ocean are the key process that controls atmosphericCO2 and not the overturning itself. However, resumption ofthe Atlantic overturning and associated injection of saltyNADW increases stratification in the Southern Ocean, lead-ing to declining CO2, less heat loss to high southern latitudesurface waters during winter and hence cooler air tempera-tures over Antarctica. These model results are qualitativelyand quantitatively consistent with observations from theglacial ice core CO2 record (Figure 1) and provide an expla-nation for their gradual changes and low frequency variability.

The simulated CO2 response is sensitive to the model back-ground climate and larger in amplitude in a warmer climatewith a stronger interstadial overturning circulation. Morework is needed to perform experiments with a more realistic


GM01073_CH20.qxd 9/8/07 5:26 PM Page 331

background state of glacial climate and carbon cycle includ-ing a realistic ice sheet cover. The latter has been shown to beimportant for the response of the terrestrial carbon inventory[Köhler et al., 2005b].

Other more idealized experiments suggest that theobserved enhancement of vertical mixing over rough topog-raphy in the Southern Ocean leads to higher atmosphericCO2 by ~40 ppmv, confirming the sensitivity of atmosphericCO2 to processes that affect stratification in Antarctic waters.The sensitivity of CO2 to mixing in the Southern Ocean alsocalls for intensified efforts to develop, improve and imple-ment process based parameterizations of mixing in ocean cli-mate models. We have shown how enhanced mixing overrough topography in the Southern Ocean affects stratificationthere and hence atmospheric CO2. We recommend the tidalmixing parameterization [Simmons et al., 2004] to be usedalso in other ocean climate models, particularly if used forbiogeochemical studies. So far most models still use moresimplified schemes. The coupled Geophysical FluidDynamics Laboratory model CM2, for example, uses theconstant vertical profile of Bryan and Lewis [1979] mimick-ing bottom intensified mixing [Gnanadesikan et al., 2006].However, other processes that lead to enhanced mixing, e.g.the interaction of the barotropic flow with rough topographyin the Southern Ocean [Sloyan, 2005], wind induced turbu-lence, or the effects of tropical cyclones on mixing in the lowlatitude thermocline [Emanuel, 2001] are much less under-stood and more basic research will be needed before para-meterisations for GCMs can be developed.

Our results appear to contrast with those of Köhler et al.[2005b] after which changes in the terrestrial carbon inven-tory cause an increase of atmospheric CO2 after a collapse ofthe Atlantic overturning. We can neither support the sugges-tion by Martin et al. [2005] after which solubility alone couldexplain the multi-millennial glacial CO2 variations. Ourresults are also in contrast to box model and other more sim-plified model simulations that show lower atmospheric CO2

in response to decreased overturning circulation [Keir, 1988;Heinze and Hasselmann, 1993; Schulz et al., 2001; Cameronet al., 2005; Köhler et al., 2005a, 2006]. The reason for thisdiscrepancy might be related to the fact that box models orother simplified models do not capture the effect of NADWon Southern Ocean stratification and sea ice. Our results areconsistent with and support the findings of Marchal et al.[1998, 1999], after which atmospheric CO2 increases after ashutdown of the Atlantic overturning due to changes in oceancarbon cycling. However, here we attribute the reason for theCO2 increase to changes in Southern Ocean stratificationsuch as deeper mixed layers and steepening of isopycnals,whereas Marchal et al. [1998, 1999] emphasize solubilitychanges.

Our working hypothesis, after which a stronger ocean cir-culation leads to higher atmospheric CO2, is apparently inconflict with the model response of higher CO2 due to a col-lapsed Atlantic overturning circulation. Obviously thishypothesis is much too simplistic to apply to the real world oreven the less complex model, but it might be reconciled withthe model results if the term “ocean circulation” refers notonly (or not mainly) to the large scale overturning circulationbut rather also includes vertical mixing processes in theSouthern Ocean which strongly control atmospheric CO2. Inresponse to a reduction of deep water formation in the NorthAtlantic, ventilation of deep and intermediate watersincreases in the other two high latitude areas of the worldocean, the North Pacific and the Southern Ocean, supportingthe ideas of Broecker [1998] and Saenko et al. [2004]. Itappears as if the renewal of global deep and intermediatewaters is resilient to perturbations and that a reduction in oneof the three deep water formation regions will be compen-sated in the other regions. Here we have shown that this canlead to reduced carbon content in the ocean as a whole.

Acknowledgments. This research has been supported by the NSFPaleoclimate Program as part of the PaleoVar project.

REFERENCES

Ahn, J., and E.J. Brook, Atmospheric CO2 and climate from 65 to 30 ka B.P.,Geophys. Res. Lett., 34, L10703, doi:10.1029/2007GL029551, 2007.

Blunier, T., and E.J. Brook, Timing of millennial-scale climate change inAntarctica and Greenland during the last glacial period, Science, 291,109-112, 2001.

Broecker, W.S., D.M. Petit, and D. Rind, Does the ocean-atmosphere systemhave more than one stable mode of operation? Nature, 315, 21-26,doi:10.1038/315021a0, 1985.

Broecker, W.S., Paleocean circulation during the last deglaciation: A bipolarseesaw? Paleoceanogr., 13, 119-121, 1998.

Brook, E.J., S. Harder, J. Severinghaus, E.J. Steig, and C.M. Sucher, On the origin and timing of rapid changes in atmospheric methane duringthe last glacial period, Global Biogeochemical Cycles, 14, 559-572, 2000.

Bryan, K., and L.J. Lewis, A water mass model of the world oceans, J. Geophys. Res., 84, 2503-2517, 1979.

Cameron, D.R., T.M. Lenton, A.J. Ridgwell, J.G. Shepherd, R. Marsh, and A.Yool, A factorial analysis of the marine carbon cycle and ocean circula-tion controls on atmospheric CO2, Global Biogeochem. Cycles, 19,GB4027, doi:10.1029/2005GB002489.

Conkright, M.E. et al. World Ocean Database 2001, Volume 1: Introduction.Edited by S. Levitus, NOAA Atlas NESDIS 42, U.S. GovernmentPrinting Office, 167 pp., 2002.

Emanuel, K., Contribution of tropical cyclones to meridional heat transportby the oceans, J. Geophys. Res., 106, 14,771-14,781, 2001.

EPICA Community Members, One-to-one coupling of glacial climate vari-ability in Greenland and Antarctica, Nature, 444, 195-198, 2006.

Ewen, T.L., A.J. Weaver, M. Eby, Sensitivity of the Inorganic Ocean CarbonCycle to Future Climate Warming in the UVic Coupled Model, Atmos.-Ocean, 42, 23-42, 2004.

Ganachaud, A. and C. Wunsch, Improved estimates of global ocean circula-tion, heat transport and mixing from hydrographic data, Nature, 408, 453-457, 2000.


GM01073_CH20.qxd 9/8/07 5:26 PM Page 332

Ganeshram, R.S., S.E. Calvert, T.F. Pedersen and G.A. Cowie, Factorscontrolling the burial of organic carbon in laminated and bioturbatedsediments off NW Mexico: Implications for hydrocarbon preservation.Geochim. Cosmochim. Acta 63, 1723-1734, 1999.

Gent, P.R., and J.C. McWilliams, Isopycnal mixing in ocean circulationmodels, J. Phys. Oceanogr., 20, 150-155, 1990.

Gnanadesikan, A. et al., GFDL’s CM2 Global coupled climate models. PartII: The baseline ocean simulation, J. Clim., 19, 675-697, 2006.

Grootes, P.M., M. Stuiver, J.W.C. White, S.J. Johnsen, and J. Jouzel,Comparison of oxygen isotope records from the GISP2 and GRIPGreenland ice cores, Nature, 366, 552-554, 1993.

Heinze, C., and K. Hasselmann, Inverse multiparameter modeling of paleo-climate carbon cycle indices, Quaternary Res., 40, 281-296, 1993.

Huber, C., M. Leuenberger, R. Spahni, J. Flueckiger, J. Schwander, T.F.Stocker, S. Johnson, A. Landais, and J. Jouzel, Isotope calibratedGreenland temperature record over Marine Isotope Stage 3 and its rela-tion to CH4, Earth and Planet. Sci. Let., 243, 504-519, 2006.

Indermühle, A., E. Monnin, B. Stauffer, T.F. Stocker, and M. Wahlen,Atmospheric CO2 concentration from 60 to 20 kyr BP from the TaylorDome ice core, Antarctica, Geophys. Res. Lett., 27, 735-738, 2000.

Jaccard, S.L., G.H. Haug, D.M. Sigman, T.F. Pedersen, H.R. Thierstein, andU. Röhl, Science, 308, 1003-1006, 2005.

Johnsen, S.J., W. Dansgaard, H.B. Clausen, and C.C. Langway Jr., Oxygenisotope profiles through the Antarctic and Greenland ice sheets, Nature,235, 429, 1972.

Key, R.M. et al., A global ocean carbon climatology: Results from GLODAP,Glob. Biogeochem. Cycles, 18, GB4031, 2004.

Keigwin, L.D., Glacial-age hydrography of the far Northwestern PacificOcean, Paleoceanogr., 13, 323-339, 1998.

Keir, R.S., On the late Pleistocene ocean chemistry and circulation,Paleoceanogr., 3, 413-445, 1988.

Köhler, P., H. Fischer, G. Munhoven, and R.E. Zeebe, Quantitative interpre-tation of atmospheric carbon records over the last glacial termination,Global Biogeochem. Cycles, 19, GB4020, doi:10.1029/2004GB002345,2005a.

Köhler, P., F. Joos, S. Gerber, and R. Knutti, Simulated changes in vegetationdistribution, land carbon storage, and atmospheric CO2 in response to acollapse of the North Atlantic thermohaline circulation, Clim. Dyn., 25,689-708, 2005b.

Köhler, P., R. Muscheler, H. Fischer, A model-based interpretation of low-frequency changes in the carbon cycle during the last 120,000 yearsand its implications for the reconstruction of atmospheric ∆14C,Geochem. Geophys. Geosys., 7, Q11N06, doi:10.1029/2005GC001228,2006.

Matsumoto, K. et al., Evaluation of ocean carbon cycle models with data-based metrics, Geophys. Res. Lett., 31, L07303, doi:10.1029/2003GL018970, 2004.

Marchal, O., T.F. Stocker, and F. Joos, Impact of oceanic reorganizations onthe ocean carbon cycle and atmospheric carbon dioxide content,Paleoceanogr., 13, 225-244, 1998.

Marchal, O., T.F. Stocker, J. Joos, A. Indermühle, T. Blunier, and J. Tschumi,Modelling the concentration of atmospheric CO2 during the YoungerDryas climate event, Clim. Dyn., 15, 341- 354, 1999.

Martin, P., D. Archer, and D.W. Lea, Role of deep sea temperature in the car-bon cycle during the last glacial, Paleoceanogr., 20, PA2015,doi:10.1029/2003PA000914, 2005.

McPhaden, M.J., and D. Zhang, Slowdown of the meridional overturning cir-culation in the upper Pacific Ocean, Nature, 415, 603-608, 2002.

Meissner, K.J., A.J. Weaver, H.D. Matthews and P.M. Cox, The role of landsurface dynamics in glacial inception: a study with the UVic EarthSystem Model, Clim. Dyn., 21, 515-537, 2003.

Naveira Garabato, A.C., K.L. Polzin, B.A. King, K.J. Heywood, and M.Visbeck, Widespread Intense Turbulent Mixing in the Southern Ocean,Science, 303, 210-213, 2004.

Neftel, A., H. Oeschger, T. Stauffelbach, and B. Stauffer, CO2 record in theByrd ice core 50,000-5000 years BP, Nature, 331, 609-611, 1988.

Orr, J.C., Najjar, R., Sabine, C.L., Joos, and F., Abiotic-HOWTO, InternalOCMIP Report, LSCE/CEA Saclay, Gif-sur-Yvette, France, 25pp., 1999.

Rashid, H., R. Hesse, and D.J.W. Piper, Evidence for an additional Heinrichevent between H5 and H6 in the Labrador Sea, Paleoceanogr., 18,doi:10.1029/2003PA000913, 2003.

Saenko, O.A., A. Schmittner, and A.J. Weaver, On the Role of Wind-Driven SeaIce Motion on Ocean Ventilation, J. Phys. Oceanogr., 32, 3376-3395, 2002.

Saenko, O.A., A. Schmittner, and A.J. Weaver, The Atlantic-Pacific Seesaw,J. Clim., 17, 2033-2038, 2004.

Saenko, O.A., and W.J. Merryfield, On the effect of topographicallyenhanced mixing on the global ocean circulation, J. Phys. Oceanogr., 35,826-834, 2005.

Sarmiento, J.L., and J.R. Toggweiler, A new model for the role of the oceansin determining atmospheric PCO2, Nature, 308, 621-624, 1984.

Sarmiento, J.L. and J.C. Orr, Three-dimensional simulations of the impact ofSouthern Ocean nutrient depletion on atmospheric CO2 and ocean chem-istry, Limnol. Oceanogr., 36, 1928-1950, 1991.

Sarmiento, J.L., Dunne, J., Gnanadesikan, A., Key, R.M., Matsumoto, K.,Slater, R., A new estimate of the CaCO3 to organic carbon export ratio,Glob. Biogeochem. Cycles, 1107, doi:10.1029/2002GB001919, 2002.

Sarmiento, J.L. and N. Gruber, Ocean Biogeochemical Dynamics, PrincetonUniversity Press, Princeton, N.J., 2006.

Sarnthein, M., K. Stattegger, D. Dreger, H. Erlenkeuser, P. Grootes, B.J. Haupt, S. Jung, T. Kiefer, W. Kuhnt, U. Pflaumann, C. Schäfer-Neth,H. Schulz, M. Schulz, D. Seidov, J. Simstich, S. van Kreveld, E.Vogelsang, A. Völker, and M. Weinelt (2001), Fundamental modes andabrupt changes in North Atlantic circulation and climate over the last 60ky – concepts, reconstruction and numerical modeling, in The NorthernNorth Atlantic: A Changing Envrionment, edited by P. Schäfer, M. Schlüter, W. Ritzrau, and J. Thiede, pp. 356-410, Springer-Verlag,Berlin.

Schartau, M., and A. Oschlies, Simultaneous data-based optimization of a1D-ecosystem model at three locations in the North Atlantic: Part I-Method and parameter estimates, J. Mar. Res., 61, 765-793, 2003.

Schmittner, A., and A.C. Clement, Sensitivity of the thermohaline circula-tion to tropical and high latitude freshwater forcing during the last glacial-interglacial cycle, Paleoceanogr., 17, doi:10.1029/2000PA000591, 2002.

Schmittner, A., Yoshimori, M. and Weaver, A.J., Instability of glacial climatein a model of the ocean-atmosphere-cryosphere system, Science, 295,1489-1493, 2002.

Schmittner, A., O.A. Saenko, and A.J. Weaver, Coupling of the hemispheresin observations and simulations of glacial climate change, Quat. Sci. Rev.,22, 659-671, doi:10.1016/S0277-3791(02)00184-1, 2003.

Schmittner, A., Southern Ocean sea ice and radiocarbon ages of glacial bot-tom waters, Earth and Planet. Sci. Let., 213, 53-62, doi:10.1016/S0012-821X(03)00291-7, 2003.

Schmittner, A., A. Oschlies, X. Giraud, M. Eby, H.L. Simmons, A globalmodel of the marine ecosystem for long-term simulations: Sensitivity toocean mixing, buoyancy forcing, particle sinking, and dissolved organicmatter cycling, Glob. Biogeochem. Cycles, 19, GB3004, 2005.

Schmittner, A., Decline of the marine ecosystem caused by a reduction in theAtlantic overturning circulation, Nature, 434, 628-633, 2005.

Schmittner, A., E.D. Galbraith, S.W. Hostetler and T.F. Pedersen, Large fluctu-ations of dissolved oxygen in the Indian and Pacific oceans duringDansgaard/Oeschger oscillations caused by variations of North AtlanticDeep Water Subduction, Paleoceanography, in press.

Schmittner, A., A. Oschlies, H.D. Matthews and E.D. Galbraith, Futurechanges in climate, ocean circulation, ecosystems and biogeochemicalcycling simulated for a business-as-usual CO2 emission scenario untilyear 4000 AD, Global Biogeochem. Cycles, submitted.

Scholze, M., W. Knorr and M. Heimann, Modelling terrestrial vegetationdynamics and carbon cycling for an abrupt climate change event,Holocene, 13, 327-333, 2003.

Siegenthaler, U. and Th. Wenk, Rapid atmospheric CO2 variations and oceancirculation, Nature, 308, 624-626, 1984.

Sigman, D.M., and E.A. Boyle, Glacial/interglacial variations in atmosphericcarbon dioxide, Nature, 407, 859-869, 2000.

Simmons, H.L., S.R. Jayne, L.C. St. Laurent, and A.J. Weaver, Tidally drivenmixing in a numerical model of the ocean general circulation, OceanModell., 6, 245– 263, 2004.


GM01073_CH20.qxd 9/8/07 5:26 PM Page 333

Sloyan, B.M., Spatial variability of mixing in the Southern Ocean, Geophys.Res. Let., 32, L18603, doi:10.1029/2005GL023568, 2005.

Toggweiler, J.R. and Samuels, B., Effect of Drake Passage on the global ther-mohaline circulation, Deep-Sea Res., 42, 477-500, 1995.

Toggweiler, J.R., Variation of atmospheric CO2 by ventilation of the ocean’sdeepest water, Paleoceanogr., 14, 571-588, 1999.

Toggweiler, J.R., A. Gnanadesikan, S. Carson, R. Murnane, and J.L.Sarmiento, Representation of the carbon cycle in box models and GCMs:1. Solubility pump, Global Biogeochem. Cycles, 17, 1026, doi:10.1029/2001GB001401, 2003.

Toggweiler, J.R., J.L. Russell and S.R. Carson, Midlatitude westerlies,atmospheric CO2, and climate change during the ice ages, Paleoceanogr,21, PA2005, doi:10.1029/2005PA001154, 2006.

Weaver, A.J., M. Eby, A.F. Fanning and E.C. Wiebe, Simulated influence ofcarbon dioxide, orbital forcing and ice sheets on the climate of the lastglacial maximum, Nature, 394, 847-853, 1998.

Weaver, A.J., et al., The UVic Earth System Climate Model: Model descrip-tion, climatology, and applications to past, present and future climates,Atmos. Ocean, 39(4), 361–428, 2001.

J. Ahn and E. J. Brook, Department of Geosciences, Oregon StateUniversity, Corvallis, Oregon, USA.

A. Schmittner, College of Oceanic and Atmospheric Sciences, 104 OceanAdmin. Bldg., Oregon State University, Corvallis, Oregon 97331, USA.([email protected])


GM01073_CH20.qxd 9/8/07 5:26 PM Page 334

impact of the ocean’s overturning circulation on...

Documents